skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Parameter Estimation for Thurstone Choice Models

Abstract

We consider the estimation accuracy of individual strength parameters of a Thurstone choice model when each input observation consists of a choice of one item from a set of two or more items (so called top-1 lists). This model accommodates the well-known choice models such as the Luce choice model for comparison sets of two or more items and the Bradley-Terry model for pair comparisons. We provide a tight characterization of the mean squared error of the maximum likelihood parameter estimator. We also provide similar characterizations for parameter estimators defined by a rank-breaking method, which amounts to deducing one or more pair comparisons from a comparison of two or more items, assuming independence of these pair comparisons, and maximizing a likelihood function derived under these assumptions. We also consider a related binary classification problem where each individual parameter takes value from a set of two possible values and the goal is to correctly classify all items within a prescribed classification error. The results of this paper shed light on how the parameter estimation accuracy depends on given Thurstone choice model and the structure of comparison sets. In particular, we found that for unbiased input comparison sets of a given cardinality,more » when in expectation each comparison set of given cardinality occurs the same number of times, for a broad class of Thurstone choice models, the mean squared error decreases with the cardinality of comparison sets, but only marginally according to a diminishing returns relation. On the other hand, we found that there exist Thurstone choice models for which the mean squared error of the maximum likelihood parameter estimator can decrease much faster with the cardinality of comparison sets. We report empirical evaluation of some claims and key parameters revealed by theory using both synthetic and real-world input data from some popular sport competitions and online labor platforms.« less

Authors:
 [1];  [2]
  1. London School of Economics (United Kingdom)
  2. Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
Publication Date:
Research Org.:
Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
Sponsoring Org.:
USDOE
OSTI Identifier:
1353026
Report Number(s):
LA-UR-17-23386
DOE Contract Number:
AC52-06NA25396
Resource Type:
Technical Report
Country of Publication:
United States
Language:
English
Subject:
97 MATHEMATICS AND COMPUTING; Mathematics

Citation Formats

Vojnovic, Milan, and Yun, Seyoung. Parameter Estimation for Thurstone Choice Models. United States: N. p., 2017. Web. doi:10.2172/1353026.
Vojnovic, Milan, & Yun, Seyoung. Parameter Estimation for Thurstone Choice Models. United States. doi:10.2172/1353026.
Vojnovic, Milan, and Yun, Seyoung. 2017. "Parameter Estimation for Thurstone Choice Models". United States. doi:10.2172/1353026. https://www.osti.gov/servlets/purl/1353026.
@article{osti_1353026,
title = {Parameter Estimation for Thurstone Choice Models},
author = {Vojnovic, Milan and Yun, Seyoung},
abstractNote = {We consider the estimation accuracy of individual strength parameters of a Thurstone choice model when each input observation consists of a choice of one item from a set of two or more items (so called top-1 lists). This model accommodates the well-known choice models such as the Luce choice model for comparison sets of two or more items and the Bradley-Terry model for pair comparisons. We provide a tight characterization of the mean squared error of the maximum likelihood parameter estimator. We also provide similar characterizations for parameter estimators defined by a rank-breaking method, which amounts to deducing one or more pair comparisons from a comparison of two or more items, assuming independence of these pair comparisons, and maximizing a likelihood function derived under these assumptions. We also consider a related binary classification problem where each individual parameter takes value from a set of two possible values and the goal is to correctly classify all items within a prescribed classification error. The results of this paper shed light on how the parameter estimation accuracy depends on given Thurstone choice model and the structure of comparison sets. In particular, we found that for unbiased input comparison sets of a given cardinality, when in expectation each comparison set of given cardinality occurs the same number of times, for a broad class of Thurstone choice models, the mean squared error decreases with the cardinality of comparison sets, but only marginally according to a diminishing returns relation. On the other hand, we found that there exist Thurstone choice models for which the mean squared error of the maximum likelihood parameter estimator can decrease much faster with the cardinality of comparison sets. We report empirical evaluation of some claims and key parameters revealed by theory using both synthetic and real-world input data from some popular sport competitions and online labor platforms.},
doi = {10.2172/1353026},
journal = {},
number = ,
volume = ,
place = {United States},
year = 2017,
month = 4
}

Technical Report:

Save / Share:
  • OASIS, a decision support system for ground water contaminant modeling, has been developed for the EPA by Rice University, through the National Center for Ground Water Research. As a decision support system, OASIS was designed to provide a set of tools which will help scientists and modelers assess ground water contamination problems. OASIS is a graphical decision support system and was developed around BIOPLUME II, a numerical model which simulates the aerobic degradation of dissolved hydrocarbons in ground water. The OASIS system was developed in the HyperCard environment and, unlike traditional software, contains extensive documentation and help onscreen. Question markmore » icons throughout OASIS lead the user to further discussion and definitions, thereby allowing the system to be fully operational without the aid of paper documentation. The information in OASIS includes documentation, a hydrogeologic database, two chemical databases, several simple hydrogeologic models, and the BIOPLUME II model with preprocessors and postprocessors. The system was developed for use on Macintosh personal computers and now contains over 1600 screens and 9 megabytes of information.« less
  • Many popular models for photovoltaic system performance employ a single diode model to compute the I - V curve for a module or string of modules at given irradiance and temperature conditions. A single diode model requires a number of parameters to be estimated from measured I - V curves. Many available parameter estimation methods use only short circuit, o pen circuit and maximum power points for a single I - V curve at standard test conditions together with temperature coefficients determined separately for individual cells. In contrast, module testing frequently records I - V curves over a wide rangemore » of irradi ance and temperature conditions which, when available , should also be used to parameterize the performance model. We present a parameter estimation method that makes use of a fu ll range of available I - V curves. We verify the accuracy of the method by recov ering known parameter values from simulated I - V curves . We validate the method by estimating model parameters for a module using outdoor test data and predicting the outdoor performance of the module.« less
  • A method is presented for the estimation of parameters in structural dynamics models. The key features of the method are as follows: estimates are obtained for the intrinsic (physical) parameters of the original model; the procedure is useful for structural dynamics problems limited to medium size, e.g., 1 to 200 degrees of freedom, where only a few, e.g., 1 to 10 key parameters are known; a constrained version of the Discrete, Iterated Extended Kalman Filter for parameter estimation is employed to obtain the estimates, utilizing frequency response function data as the measurements; a slight variation of modal synthesis transformation methods,more » including reduction of boundary coordinates, is developed to reduce the order of the original model when the frequency response function and its associated partial derivatives are needed for the estimation algorithm; and the magnitude of the frequency response function /H/sub ..cap alpha beta../(..omega..)/ is used as data for the estimation procedure. A detailed development of the filter and transformation equations is presented, followed by a discussion of the measurement noise characteristics for the practical case, as well as the simulated data used in the numerical study. Results are given for the several cases of an example problem, demonstrating the effectiveness of the method. Following the conclusions are suggestions for future research.« less
  • The purpose of this project is to estimate dispersion in the presence of heavy RF background noise and then to find wide, band events.